The Fraunhofer IML's Future Congress, taking place on September 12th and 13th, is themed "Logistics goes AI." Participating organizations include IML and Digital Hub Logistics, as well as the Lamarr Institute for Machine Learning and the Open Logistics Foundation. Successful models such as Julian Hinxlage's "CV on Edge" will be presented.
AI-based image capture and processing – computer vision, or CV for short – enables the tracking of goods and the measurement of load carriers using cameras. The Silicon Economy, with its "CV on Edge" project, is among the pioneers of this new technology. Researchers are developing a complete package consisting of a smart camera and an AI-based, do-it-yourself image processing service. All developments are released as open source – so that they can quickly spread and become established in the logistics sector, especially among small and medium-sized enterprises.
In this interview, Silicon Economy researcher Julian Hinxlage, research associate at Fraunhofer IML and Product Owner of «CV on Edge», explains what contribution software and hardware can make to greater efficiency in goods receipt, pallet management or at the warehouse.
How does logistics benefit, or which logistics areas benefit, from computer vision?
Julian Hinxlage: AI-based image capture and processing with cameras makes it possible to detect, classify, and identify objects. There are countless applications for this in logistics. Just one example: Most companies have no overview of where empty containers are located in their warehouses. They're simply left lying around, and forklift drivers spend a lot of time searching for and finding them. Computer vision provides this overview: Cameras are installed in the warehouse to capture images of all containers. Artificial intelligence can then be used to locate empty containers. Currently, this works exceptionally well indoors – but with the "Yard Lens" use case we're currently implementing, we're also expanding to outdoor areas.
Artificial intelligence, that sounds complicated…
No, it's not rocket science. With our Guided Training Service, any company can train an artificial intelligence – with just a few settings and steps, and depending on the complexity, in just a few minutes. This is a huge innovation. A company only needs to upload a few images of the object to be recognized or counted to the service: 20 or 30 photos are enough for basic detections! The more images uploaded, the more robust the application becomes. With just a few images, the AI recognizes containers during the day, and with more images, it can also recognize them at night. The more sophisticated the architecture and the denser the neural networks, the more accurately the AI recognizes its target objects.
Fig.: IML
The service facilitates the use of MLCVZoo – a component developed in the Silicon Economy for image processing applications…
MLCVZoo combines different machine learning algorithms at the program code level and is therefore geared more towards ML developers. Building on this, we've created "CV on Edge," a do-it-yourself service that is self-explanatory and intuitive. No IT expertise is required, just a basic understanding.
Is the service already available?
Like all developments from the Silicon Economy, we will soon make the software available as open source. We are also continuously developing the service and already have many ideas. For example, we want to enable users to store different models in a single database. A company can then use an existing model and supplement it with its own individual images to adapt it to its specific needs or make it more robust.
How great is the interest from companies?
Very big! This year we've already had the opportunity to present the service to companies at major industry trade fairs. The feedback has been very positive. Many can immediately think of a possible use case: from counting containers to recognizing hazardous materials labels. Our solution is also very interesting for logistics IT service providers. We know that many are already waiting for our official release…
The smart camera from «CV on Edge» seems a major selling point to be. What's so special about this camera?
The smart camera is a so-called edge camera, meaning that the intelligence and data processing take place directly on the camera. This minimizes the amount of data that needs to be transferred through the network or via mobile networks. Furthermore, the protection of personal data is guaranteed. Because no infrastructure is required beyond Wi-Fi and a power source, the camera can be used almost anywhere. During development, we made sure that costs remained reasonable: they range between €400 and €1,000 per unit. Thanks to modular, commercially available components, different parts can also be used within the housing.
Will the building plan also be released as open source?
Absolutely. With these building plans, you can simply print out the camera – and get started right away.
